On Sun, Jan 25, 2004 at 11:15:17AM +0200, Tzafrir Cohen wrote: > On Sun, Jan 25, 2004 at 10:30:15AM +0200, Muli Ben-Yehuda wrote: > > On Sun, Jan 25, 2004 at 10:21:29AM +0200, Tzafrir Cohen wrote: > > > > > >From looking at the source I could only ssee that it means "something is > > > fishy". > > > > 0 order allocation failed means the kernel couldn't allocate even one > > page of memory. Unless the machines are pretty much out of memory, > > that should never happen. > > > > > In one case replacing reiserfs with ext3 made the problem go away. > > > Naturally this is a drastic solution that I don't want to take. > > > > > > Anybody seen this lately? > > > > Actually, yes, on lkml, unless my memory is playing tricks on me. But > > I don't remember the details, sorry. Try the archives... > > Makes some sense. I did see a number of huge processes (e.g: tar eating > out almost all of my memory, and likewise "proxymap" of postfix). > > I'm positive that the tar process was run as non-root without any > privilges. > > I also saw VM killing messages. A number of processes managed to get > killed around the time I saw that message.
Interesting development here: THe system has been upgraded from Mandrake 9.0 to Mandrake 9.2 using urpmi . It went generally smoother than I thought, but troubles began immidietly after that. I had not changed the kernel, so I initially figured it was some new kernel module being loadd for the first time or so. After a bit of stracing on two such "inflating" processes (tar and devfsd) I noticed that they were in a countinous loop always opening , reading-form and closing /etc/group. And they kept allocating more memory. I compared the existing /etc/group to the new defaault one (/etc/group.rpmnew) and noticed that: 1. my original file had 'x' in the password field, whereas the original had this field empty 2. The original had a group with many members at the last line. After just deleting the 'x'-s devfsd seems to manage to work. Tar still looped, so I removed the big group, and it functioned as usual. So the imidiate suspect is glibc's handling of edge cases. Current glibc: glibc-2.3.2-14mdk I hope his works > > > > > Which kernel is this with? > > > > This time: 2.4.23-pre6 . But I also had something similar with 2.4.23 > > And to Shachar: thanks, I tryed that. But the corruption re-appeared too > soon :-( > > -- > Tzafrir Cohen +---------------------------+ > http://www.technion.ac.il/~tzafrir/ |vim is a mutt's best friend| > mailto:[EMAIL PROTECTED] +---------------------------+ > > ================================================================= > To unsubscribe, send mail to [EMAIL PROTECTED] with > the word "unsubscribe" in the message body, e.g., run the command > echo unsubscribe | mail [EMAIL PROTECTED] -- Tzafrir Cohen +---------------------------+ http://www.technion.ac.il/~tzafrir/ |vim is a mutt's best friend| mailto:[EMAIL PROTECTED] +---------------------------+ ================================================================= To unsubscribe, send mail to [EMAIL PROTECTED] with the word "unsubscribe" in the message body, e.g., run the command echo unsubscribe | mail [EMAIL PROTECTED]